2 min read AI-generated

Z.ai GLM-5.2: 753 Billion Parameters, Open Weights, MIT License

Copy article as Markdown

Z.ai releases GLM-5.2 — an open-weights coding model with 1 million token context that claims to beat GPT-5.5 on coding benchmarks.

Featured image for "Z.ai GLM-5.2: 753 Billion Parameters, Open Weights, MIT License"

The open-source AI scene gets fresh competition from China: Z.ai has released GLM-5.2, a 753-billion-parameter model under the MIT license, optimized specifically for autonomous coding and engineering tasks.

The specs

GLM-5.2 ships with a 1-million-token context window and is available for free download on Hugging Face. Companies can run it locally, fine-tune it, and integrate it into their own infrastructure — no license restrictions.

Z.ai also offers the model via its own API and over 20 coding environments. Enterprise subscriptions start at $12.60 per month — a fraction of what proprietary alternatives charge.

What makes it technically interesting

The most notable innovation is called ‘IndexShare’: an architectural optimization that reuses the same indexer across every four sparse attention layers. This cuts per-token compute at maximum context length by a factor of 2.9. For long coding sessions with large codebases, that’s a real advantage.

The benchmark question

Z.ai claims GLM-5.2 beats GPT-5.5 on multiple long-horizon coding benchmarks — at one-sixth the cost. Whether that holds up in practice remains to be seen. Chinese AI companies are known for aggressive benchmark claims that don’t always translate to real-world performance.

Still: an open-weights model of this scale under the MIT license is a serious offering. Especially for companies that want to run coding agents locally — whether for data privacy or to stay independent from API providers.

The bigger picture

The trend is clear: open-source models are catching up. MiniMax M3, Kimi K2.7-Code, now GLM-5.2 — the gap to proprietary frontier models keeps shrinking. That’s good for competition and good for developers who want choice.

For Anthropic and OpenAI, this means: pricing pressure from below is growing. If an open-weights model at $12.60 a month handles 80 percent of tasks just as well, justifying premium prices gets harder.


Sources: